International Journal of Pharmaceutics 453 (2013) 641647

Contents lists available at SciVerse ScienceDirect

International Journal of Pharmaceutics

journal homepage: www.elsevier.com/locate/ijpharm

Pharmaceutical nanotechnology

Using machine learning for improving knowledge on antibacterial

effect of bioactive glass
M.M. Echezarreta-Lpez , M. Landin
Departamento Farmacia y Tecnologa Farmacutica, Facultad de Farmacia, Campus Vida, Universidad de Santiago, Santiago de Compostela 15782, Spain

a r t i c l e i n f o a b s t r a c t

Article history: The aim of this work was to nd relationships between critical bioactive glass characteristics and their
Received 21 March 2013 antibacterial behaviour using an articial intelligence tool. A large dataset including ingredients and
Received in revised form 13 June 2013 process variables of the bioactive glasses production, bacterial characteristics and microbiological exper-
Accepted 15 June 2013
imental conditions was generated from literature and analyzed by neurofuzzy logic technology. Our
Available online 24 June 2013
ndings allow an explanation on the variability in antibacterial behaviour found by different authors and
to obtain general conclusions about critical parameters of bioactive glasses to be considered in order to
Keywords:
achieve activity against some of the most common skin and implant surgery pathogens.
Bioactive glass
Antibacterial behaviour 2013 Elsevier B.V. All rights reserved.
Modelling
Articial intelligence
Neurofuzzy logic
Machine learning

1. Introduction prevention of potential postoperative infections. Different authors

Since the introduction by Hench in the 1970s of Bioglass
(Hench et al., 1971; Hench and Paschall, 1973) this material has
been extensively investigated. Bioactive glasses (BG) are special
glass systems composed by SiO2 , CaO, P2 O, and NaO that can be pro-
duced by traditional melting or by solgel processes (Hench et al.,
1971; Hench, 1998a). Their bioactive behaviour comes from their
ability of bonding soft and hard tissues through complex reactions,
which form strong and compliant interfaces between the glass and
the tissues (Hench et al., 1971). BGs have excellent biocompatibil-
ity, osteoconductivity and osteostimulation (Sun et al., 2007). For
these reasons BGs have been employed for clinical use for a variety
of medical applications, including surgical orthopaedics and den-
tistry, mainly for repairing osseous, cystic, tumours and periodontal
defects, or other lesions sites after resection in the appendicular
skeleton (Hench, 1998b; Kellomkiki et al., 2000; Vogel et al., 2001;
Froum et al., 2002).
The use of implants in the body is associated with a risk of
bacterial colonization of the material and the subsequent failure
of the surgery (Gristina, 1987; Pye et al., 2009; Campoccia et al.,
2006). Combinations of implant/drugs or implant materials with
antibacterial properties would represent an excellent approach for
Corresponding author. Tel.: +34 881815252; fax: +34 881815038.
E-mail addresses: mmagdalena.echezarreta@usc.es, ffmagda@yahoo.es
(M.M. Echezarreta-Lpez).
have pointed out the utility of the antibacterial activity of BGs in
oral (Allan et al., 2001; Zehnder et al., 2006; Waltimo et al., 2009),
orthopaedic implants (Xie et al., 2009; Gorriti et al., 2009; Misra
et al., 2010) and wound dressings (Verrier et al., 2004; Day and
Boccaccini, 2005).
A review of the literature allows the conclusion that antibac-
terial properties of BGs depend on their composition and
morphological characteristics (Hu et al., 2009; Kalmodia and Molla,
2010), but the wide variety of conditions used to perform the
studies (e.g. bacteria studied, antimicrobial conditions or glass
pretreatment) hinders the specic factors inuencing bactericidal
capacity and the best combination of characteristics in achieving
the maximum antibacterial effect.
The analysis of a database generated from historical data is dif-
cult using traditional approaches. However, articial intelligence
tools allow for this, and the discovery of general or new patterns
from this type of data in a process named data mining (Rowe and
Colbourn, 2003). Among all methods, the neurofuzzy logic (NFL)
approach has proven its utility in modelling complex non-linear
relationships hidden in data, having a higher accuracy in predic-
tion than classical statistics and helping the understanding of the
complex relationships between variables (Shao et al., 2006; Landin
et al., 2009; Gallego et al., 2011).
NFL combines the adaptive learning capabilities from articial
neural networks (ANN) with the generality of representation from
fuzzy logic through simple rules. It has been demonstrated that
ANN is effective in modelling complex processes. However, the

0378-5173/$ see front matter 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.ijpharm.2013.06.036
642 M.M. Echezarreta-Lpez, M. Landin / International Journal of Pharmaceutics 453 (2013) 641647
interpretation of an ANN is not always easy, especially if an impor-
tant number of inputs are involved because just black box models
are generated. For that reason other technologies such fuzzy logic
are being coupled to ANN to help handling those complex mod-
els, which generates, after a proper fuzzication process of the
variables, simple IF. . . THEN rules (Garca-Infante et al., 2010).
Neurofuzzy logic can be applied to historical databases generat-
ing understandable and reusable knowledge in an explicit format.
(Rowe and Landn, 2013)
We think that neurofuzzy logic can be useful in establish-
ing the critical variables (composition, variables involved in the
BGs production and/or variables used to demonstrate antibacterial
properties) to detect antibacterial effect of BGs.
In this paper we have reviewed the literature on antimicrobial
activity of bioactive glass, to extract and analyze together previous
results. We have modelled results using neurofuzzy logic, intending
to establish the critical aspects in the determination of the antimi-
crobial activity of the BGs.
2. Materials and methods
A large database (531 facts) on antimicrobial properties of sev-
eral bioactive glasses was generated from 10 previously published
articles from 2000 to 2010 (see table at the supplementary mate-
rial) (Mortazavi et al., 2010; Zhang et al., 2010; Hu et al., 2009;
Leppranta et al., 2008; Munukka et al., 2008; Waltimo et al., 2007;
Zehnder et al., 2006; Abou Neel et al., 2005; Bellantone et al., 2000,
2002).
In general, the antimicrobial activity studies intend to probe
the antibacterial effect of BGs by culturing the pretreated or non-
pretreated biomaterials (variable in composition, physical chemical
properties and concentration) together with selected bacteria sus-
pensions in specic culture media. After a preset time of cultivation,
the pH is measured and a volume of the culture is transferred,
cultivated on a new media and the growth index (GI) measured
(Munukka et al., 2008; Mortazavi et al., 2010).
Two models were carried out. For model A, variables selected
(21 inputs) which reduced the dataset to 348 facts can be classied
in three groups: BGs characteristics, bacterial characteristics and
microbiological experimental conditions.
Among the BGs characteristics, the production method, BGs
composition regarding compound concentration (SiO2 , CaO, Na2 O,
P2 O5 , MgO, K2 O, Al2 O3 , B2 O3 , Ag2 O, CuO) in the bioactive glass
(w/w, %), particle size as mean size (m), morphology as nanopar-
ticles (nano), powder (po), particles (pa) and bres (b), and BG
concentration (mg/mL) were registered.
For every experiment bacterial characteristics recorded were:
microorganism specie including Staphylococcus aureus (Sa), Staphy-
lococcus epidermidis (Se), Enterococcus faecalis (Ef), Escherichia coli
(Ec), Pseudomonas aeruginosa (Pae) and Salmonella typhi (Sty), and
morphology: bacillus (Ba) or coccus (Co). Bacteria were grown
aerobically in different microbiological experimental conditions
registered as bacterial concentration (CFU/mL), culture time (h) and
culture media categorized in two groups named as buffered solu-
tions (BS) (when used, simulated buffered uid, phosphate buffered
saline or solution saline) and nutrient solution (NS) (when used,
Tryptone Soy broth, Lysogeny broth, Mueller Hinton broth or nutri-
ent broth). Additionally, coculture time and coculture media (using
the same categorization as previously noted) were registered.
For model B (Fig. 2), recorded nal pH values were also included
as an additional input. This approach reduced the number of facts
to 129.
For both, model A and B, the output recorded was the growth
index, measured as a function of the number of survival bacteria
colonies. Absence of growth (GI = 0) indicates bactericidal effect.
Table 1
The training parameters setting with FormRules 3.31 for models A and B.
Minimization parameters
Ridge regression factor: 1 e6
Model selection criteria
Structural risk minimization (SRM)
Number of set densities: 2
Set densities: 2, 3
SRM parameters: C1 = 1 and C2 = 4.8
Adapt nodes: TRUE
Max. inputs per submodel: 4
Max. nodes per input: 15
Very sparse, sparse and moderate values of GI (GI = 1 between 0
and 5 colonies, GI = 2 between 5 and 50, and GI = 3 between 50 and
300 colonies, respectively indicate moderate growth. The growth
index of 4 (>300 colonies) indicates no effect.
Database with the inputs and the output from the sources are
compiled in Table 1 of supplementary material.
2.1. The software tool: neurofuzzy logic
A commercial neurofuzzy logic software, FormRules v3.31
(Intelligensys Ltd., Stokesley, UK) was used.
The training process was conducted, in the same way, as
reported previously by Shao and coworkers (2006). The accuracy of
the neurofuzzy logic models was assessed using the ANOVA param-
eters and correlation coefcient (R2 ) for the output growth index.
n 2
(yi yi )
R2 = 1 i=1
n 2
(1)
i=1
(yi yi )
where y is the mean of the dependent variable, and y is the predicted
value from the model. Values of correlation coefcient between
70 and 99.9% together with ANOVA f values over its critical val-
ues for the corresponding degrees of freedom are an indication of
reasonable model predictabilities (Colbourn and Rowe, 2009).
Within the statistical tness criteria included in FormRules 3.31
software, structural risk minimization (SRM) was selected to give
the model with the best correlation coefcient and, simultaneously,
the simplest and more intelligible rule set. Selected parameters for
training are presented in Table 1.
3. Results
The common factor in the published studies on antibacterial
properties of BGs of the literature is the great variability in the con-
ditions used with regard to the bacteria types, compositions and
sizes of BGs as well as microbiological parameters used. In this situ-
ation to establish general conclusions on the optimal BGs properties
to assure antibacterial activity is a complex task.
3.1. Model A characteristics
The 21 inputs selected for model A and the corresponding
growth index reect the main variables (independent and depend-
ent) included in the articles from literature which made it possible
to generate a database that can be modelled by neurofuzzy logic
tool. A large database of 348 facts on antibacterial properties of BGs
materials was successfully modelled (train set r2 = 76.31) by neuro-
fuzzy logic technology (see ANOVA model in Table 2) that allowed
the selection of the critical factors or inputs which accurately
explain the variability of the growth index, selected as output.
Fig. 1 presents the four submodels considered by neurofuzzy
logic to explain variability in the GI and the inputs included by
them. It is interesting to note the reduced number of inputs, ve
 M.M. Echezarreta-Lpez, M. Landin / International Journal of Pharmaceutics 453 (2013) 641647 643

Fig. 1. Graphical example of model A with the signicant submodels and their domains developed by neurofuzzy logic for the output GI. (A) Simplest submodels generated
pointing out the signicant input interaction of SiO2 and BG concentration (in purple) as the main effect and the interaction of BG concentration and bacteria specie, the single
effects of Ag2 O and bacteria concentrations affecting growth index; (B) domains established by neurofuzzy logic technology for the continuous variables BG concentration,
Ag2 O and bacteria concentration; (C) domains established for the continuous input SiO2 . (For interpretation of the references to color in this gure legend, the reader is
referred to the web version of the article.)

(SiO2 and Ag2 O concentration, BG concentration, bacteria specie Neurofuzzy logic technology allows a set of IF . . . THEN
and bacteria concentration) out of the twenty one that neurofuzzy rules per submodel with their corresponding membership degrees
logic has selected as the ones having important effect on the growth (Table 3) to be generated.
index or, in other words, enough to explain the GI variability. Those rules should be read as follows:

Fig. 2. Graphical example of model B with the signicant submodels and their domains developed by neurofuzzy logic for the output GI. (A) Simplest submodels generated
pointing out the signicant input interaction of Si concentration and BG concentration as the main effect and the interaction of calcium concentration and bacteria specie,
the single effect of pH; (B) domains established by neurofuzzy logic technology for the continuous input pH; (C) domains established for the continuous variables BG
concentration, SiO2 and calcium concentration.
644 M.M. Echezarreta-Lpez, M. Landin / International Journal of Pharmaceutics 453 (2013) 641647

Table 2
Analysis of variance for the model A developed. Computed f ratio is higher than 1.58 which is the critical f value (19 and 348 degrees of freedom and < 0.01) to assure
statistical signicance of the model A.

Source of variation Sum of squares Degrees of freedom Mean squares Computed f ratio

Model 527.8 19 27.78 55.81

Error 163.7 329 0.49
Total 691.6 348

For example, for Rule 1 of Table 3: IF the concentration of BG
used is LOW AND the concentration of silicon oxide in the BG is
LOW THEN the Growth Index founded is LOW with a condence
level or membership of 1.00.
The meaning of LOW, MID and HIGH for each parameter has
been represented in Fig. 1B and C, which show the relationship
between input values (x-axis) and the corresponding membership
(y-axis) (Landn et al., 2009; Gallego et al., 2011; Colbourn and
Rowe, 2009; Kavli and Weyer, 1994). Using those rules, neuro-
fuzzy generates understandable and reusable knowledge on the
antibacterial activity of BGs.
3.2. Model B characteristics
A second model was carried out including 22 inputs (21 from
model A plus pH media value) and GI as the only output. The
reduce number of facts, including a measured pH value, gives a
reduced database (129 facts). Model predictability was higher than
that obtained for model A with a training R2 equal to 81.64 and
statistically signicant with a computed f ratio of 28.79 ( < 0.01).
Despite the models pointing out different inputs the informa-
tion reported by them is complementary, their main differences
being interesting analyzed. Firstly, for model B (Fig. 2), pH (the
new included variable) by itself is one of the parameters that
explains the variability of GI. Additionally, the submodel 2 of model
B includes the interaction between bacteria specie and calcium con-
centration instead of bacteria specie and BG concentration included
in model A. This interaction is the most important, explaining the
variability (purple lines) and helps in understanding the mecha-
nism of antibacterial activity of these biomaterials. Finally, the Ag2 O
concentration could not be included as a signicant input in model
B because its variability in the 129 facts is null, therefore this model,
misses the differences regarding the BG inclusion of Ag ions pointed
out by model A. Table 4 includes the rules derived from model B.
4. Discussion
Preventing and/or limiting bacteria colonization and/or con-
taminations when a biomaterial is introduced into the body, is
the main concern of a number of studies over the recent years.
The use of combined drug-materials, and especially biomaterials
showing antimicrobial activity is bound up with a lower probabil-
ity of surgery failure by infections and an increase of application
areas. Among biomaterials, bioactive glasses of specic compo-
sitions have exhibited antibacterial activity whose identication
and characterization have been demonstrated in a large range of
experimental conditions that make general conclusions difcult to
achieve.
The inclusion of data from different works and their modelling
using articial intelligence methods should allow those general
conclusions to be established.
Articial neural networks (ANNs) analysis has proved useful in
nding the relationship between the BGs solubility behaviour and
their chemical composition and if using a database large enough it
may be possible to predict their solubility with sufcient accuracy
(Brauer et al., 2007). ANNs have also been applied to characterize

Table 3
Rules for growth index generated by neurofuzzy logic for model A. Poor combinations have been highlighted.

Rule [BG] Si2 O Ag2 O Bacteria specie Bacteria concentration GI Condence level

Submodel 1
1 LOW LOW LOW 1.00
2 LOW MID HIGH 1.00
3 LOW HIGH HIGH 1.00
4 IF HIGH LOW THEN LOW 1.00
5 HIGH MID LOW 1.00
6 HIGH HIGH HIGH 1.00

Submodel 2
7 LOW HIGH 1.00
8 HIGH LOW 1.00

Submodel 3
9 LOW E. faecalis HIGH 1.00
10 HIGH E. faecalis HIGH 0.58
11 LOW S. epidermidis LOW 1.00
12 HIGH S. epidermidis LOW 1.00
13 LOW S. typhi LOW 0.88
14 IF HIGH S. typhi THEN HIGH 0.59
15 LOW E. coli LOW 0.67
16 HIGH E. coli LOW 1.00
17 LOW S. aureus LOW 0.72
18 HIGH S. aureus LOW 1.00
19 LOW P. aeruginosa LOW 0.90
20 HIGH P. aeruginosa LOW 0.68

Submodel 4
21 IF LOW THEN HIGH 0.97
22 HIGH LOW 0.72
 M.M. Echezarreta-Lpez, M. Landin / International Journal of Pharmaceutics 453 (2013) 641647 645

Table 4
Rules for growth index generated by neurofuzzy logic for model B. Poor combinations have been highlighted.

Rule pH CaO Bacteria specie [BG] Si2 O GI Condence level

Submodel 1
1 LOW HIGH 1.00
2 IF MEDIUM THEN HIGH 1.00
3 HIGH LOW 1.00

Submodel 2
4 LOW E. faecalis LOW 0.55
5 HIGH E. faecalis LOW 0.70
6 LOW S. epidermidis LOW 0.84
7 HIGH S. epidermidis LOW 0.89
8 LOW S. typhi LOW 1.00
9 HIGH S. typhi HIGH 1.00
10 IF LOW E. coli THEN LOW 1.00
11 HIGH E. coli LOW 1.00
12 LOW S. aureus LOW 1.00
13 HIGH S. aureus LOW 1.00
14 LOW P. aeruginosa LOW 1.00
15 HIGH P. aeruginosa HIGH 1.00

Submodel 3
16 LOW LOW HIGH 1.00
17 LOW HIGH HIGH 1.00
18 HIGH LOW LOW 1.00
19 HIGH HIGH HIGH 1.00

the antimicrobial activity of a heterogeneous group of compounds
and differentiate between active and inactive compounds (Garcia-
Domenech and Ortiz, 1998) or to predict the antibacterial activity
of a series of imidazole derivatives (Domingues et al., 2004).
In the present work we have used neurofuzzy logic technology
which is an articial intelligence tool that combines the adaptive
learning capabilities from ANNs with the generality and the exibil-
ity of representation from fuzzy logic allowing the growth index to
be modelled (GI) as a function of the big set of parameters included
as inputs. For the model A, using the whole dataset, neurofuzzy
logic has selected 348 facts and GI was modelled using just 5 out
the 21 parameters included.
Differences in BGs antimicrobial activity, regarding its produc-
tion method have been extensively referenced in the literature, the
solgel method being the best in producing high porosity particles
that promote higher ion release to the medium (Mortazavi et al.,
2010; Peltola et al., 1999).
Different authors have also stated that particle size is deter-
minant when BG is showing antibacterial activity. Reduction in
particle size, and as a result, the increase in the surface area of
the material, has been pointed out as an approach of increasing
the antibacterial activity of BGs (Gorriti et al., 2009; Misra et al.,
2010; Waltimo et al., 2007; Balamurugan et al., 2008). Despite these
arguments, neurofuzzy logic technology was unable to nd even
subtle effects of production procedure, particle size or morphology
when the data are analyzed together, meaning that they cannot
explain the variations in GI found from different authors in various
experiments.
The main effect explaining the variability of growth index is the
interaction between the silicon concentration and the concentra-
tion of BG (Table 3, model A, submodel 1) in the experiment, the
BG having no effect when its concentration is low and the silicon
concentration is medium or high (submodel 1, rules 2 and 3) or the
BG concentration is high and the silicon concentration is also high
(submodel 1, rule 6) (highlighted text). It has been shown that the
antibacterial action of a BG can be affected by its chemical compo-
sition and its dissolution properties in the surrounding medium. In
an aqueous environment, reactions occur on the BG particles sur-
face, including release of soluble silica, sodium and calcium which
result in an increase of the aqueous pH value (Hench, 1998a).
Our ndings agree with previous authors which correlate BG
antibacterial properties with the immediate release of alkaline
species (Waltimo et al., 2007). The higher the silicon concentration
the lower the alkaline species that can be release to the medium and
the lower its bactericidal effect. The BGs concentration increases
the total amount of ions available to be dissolved.
Some BGs, without specic ions, have shown antibacterial inhi-
bition by attaching directly the bacteria by a biologically active
carbohydroxyapatite (CHA) layer, a mechanism related to physical
interactions more than chemical interactions (Stoor et al., 1999;
Vaahtio et al., 2006; Stoor et al., 1998). However, the incorpora-
tion of particular ions into the silicate networks such silver, cooper
or boron has been specically investigated with the aim of devel-
oping antibacterial materials with promising results (Gerhardt and
Boccaccini, 2010; Bellantone et al., 2002; Mulligan et al., 2003). Pos-
sible mechanisms for bacterial inhibition by those elements have
been related to the disruption of bacteria functionality; the interfer-
ence with electron transport, the binding to DNA or the interaction
with the cell components (Balamurugan et al., 2008).
From our results from model A (Table 3), it can be deduce that
silver is decisive for the BG antibacterial activity (submodel 2, rule
8) at the concentrations included in the study. Its effect is con-
rmed when the Ag2 O concentration is higher than 1.5%, which is
in agreement with data from literature (Balamurugan et al., 2008;
Bellantone et al., 2002). Vern et al. (2008) have found that antimi-
crobial effect of silver based BGs is due to the leaching of silver from
the glass matrix, the change in pH and the ionic strength.
Different authors have related the BG antibacterial effect to
the increase of aqueous pH value caused by the release of alkali
ions from BG particles (Allan et al., 2001; Stoor et al., 1998). In
general, all articles reviewed consider that the variation environ-
mental pH is a critical parameter to the antimicrobial activity of
bioactive glasses; Mortazavi et al., 2010; Bellantone et al., 2002;
Zhang et al., 2010). Our results support those ndings. When pH is
included as input (model B) despite the number of facts being dra-
matically reduced (129) an acceptable model was produced. This
simplest model missed the ability of detecting variations due to
Ag ion concentration that remain constant in the smaller database
but succeeded explaining the GI as a function of pH variability. The
relationship silicon and nal concentration is also clearly pointed
out together with the signicant effect of the interaction between
calcium ion concentration and bacteria specie.
The database generated included six species of bacteria, three
coccus and Gram-positive (S. aureus, S. epidermidis and E. faecalis)
646 M.M. Echezarreta-Lpez, M. Landin / International Journal of Pharmaceutics 453 (2013) 641647
and three bacillus and Gram-negative (E. coli, P. aeruginosa and S.
typhi) which provided a structural and morphological variety, with
different response against physicalchemical and environmental
conditions (Allan et al., 2001; Munukka et al., 2008). Most of these
are oral bacteria, including those associated with caries and peri-
odontal diseases.
The susceptibility of species to antimicrobial agents varies
regarding the presence or absence of a lipopolysaccharide layer
(Gram-negative or Gram-positive) in their cellular wall and their
kind of association and growth, or their biolm formation capac-
ity which is different for coccus and bacillus (Mortazavi et al.,
2010; Zhang et al., 2010; Nikaido, 1993). In this study both char-
acteristics were evaluated together as one input because all the
coccus selected were Gram-positive and all the bacillus were Gram-
negative, but neurofuzzy logic did not show this input as signicant
in explaining the variation of GI.
However, as it can be derived from model A (Table 3), sub-
model 3 (rules 920), which shows the interaction between the
total BGs concentration and the bacterial specie, the susceptibility
of the different species to the BGs treatment has great variability
in condence levels or membership values, E. faecalis and S. typhi
being the most resistant species to treatment with BGs (highlighted
text).
Those differences can be explained by variations in the structure
or association of bacteria. Among the coccus studied the E. faecalis
has been described to have associations by couples forming short
chains. However, both S. epidermidis and S. aureus produce clusters.
Those differences could justify variations in the susceptibility of
these species (Nikaido, 1993).
S. typhi differs from the other two bacillus (E. coli and P.
aeruginosa) in the structural organization of the lipopolysaccharide
layer which gives bacteria resistance against BG.
As we have pointed out previously, model B (Table 4) helps to
corroborate the differences in behaviour for the species studied and
to understand the antibacterial mechanism of BG related to calcium
ion variations. Submodel 2 (rules 415) species the signicant
interaction between those inputs. In general, the coccus (Gram-
positive) species selected are more susceptible to treatment than
bacillus (Gram-negative). For the coccus group, the higher the BG
calcium concentration is the more important its antibacterial effect
is. Comparing membership values it can be concluded that among
coccus, E. faecalis can be pointed out as the most resistant specie to
treatment (GI LOW but with memberships of 0.55 and 0.70), which
is in agreement with conclusions from model A. Calcium ions are
lethal for bacteria, but E. faecalis is able to survive in calcium ion
rich environments (Nakajo et al., 2006). McHugh and coworkers
evaluated the pH required to inhibit growth in vitro and showed
that a pH higher than 11 is needed to eliminate this microorganism
(McHugh et al., 2004). Nakajo group (Nakajo et al., 2006) suggested
that the low effect of the pH increment may also be attributed to
the resistance of the cytoplasmic membrane against acid or alkaline
media along with the transport system ATP-linked proton.
The bacillus selected (all Gram-negative) have a thinner layer
of peptidoglycan and an outer membrane. The outer membrane
is a lipid bilayer and constitutes the major barrier permeability
for this group of bacteria. The lipid bilayer seems similar to any
biological membrane, but its structure is asymmetric and its com-
position differs from other living organisms. The semi-internal
layer is comprised of phospholipids, and the external layer by a
lipopolysaccharide (LPS). The stability of the LPS layer depends on
the calcium concentration. Agents and/or processes that disturb
the calcium ion equilibrium in the media can affect bacterial struc-
ture and as a consequence their integrity and bacteria mortality
(Nikaido, 1993, 2003). In fact, controversy in the use of calcium
hydroxide as an antimicrobial agent can be found in the den-
tistry literature (Siqueira and Uzeda, 1997; Estrela et al., 2001). Hu
and coworkers have shown remarkable differences between Gram-
negative (more sensitive to calcium ions) and Gram-positive (Hu
et al., 2009).
Different functions have been described for calcium ions in
Gram-negative bacteria depending on the amplitude and duration
of the calcium transient and the physiological state of the organism;
indispensable triggers like in chemostaxis, modulators of a specic
function and structural elements of the membranes, nucleoids and
spores (Norris et al., 1996).
Calcium ions act as a membrane stabilizer for S. typhi and S.
aeruginosa. Calcium ions cannot be produced or destroyed by the
bacteria, thus, bacteria should take calcium ions from the extracel-
lular media which acts as the stabilizer, increasing bacteria survival
and as a consequence reducing BGs bactericidal effect (Elliott,
2001).
Some Gram-negative species have membranes that can include
organic and inorganic ion transporters with homeostatic functions
able to capture or expelled specic ions (Jimenez and Cervantes,
2006). E. coli have specic transporters (related to nutrients active
transport) to capture calcium ions which (Onoda et al., 1992).
Osmotic stress produced by the increase of calcium in the media
alters this function reducing the viability of E. coli (Roth et al., 1985).
The presence of those specic channels justies the differences in
sensitivity of E. coli with regard to P. aeruginosa and S. typhi.
Neurofuzzy logic did not point out the inuence of any of the
microbiological parameters studied meaning that as far as the
microbiological conditions (media and time of culture and co-
culture), are adequate for a particular specie, those variables have
no effect on growth index obtained.
5. Conclusions
Articial intelligence technology allowed a novel and integrated
analysis of the results from literature on antibacterial activity of
bioactive glasses. Neurofuzzy logic was able to model an extensive
database on bioactive glasses, to determine the critical variables for
BG antibacterial activity and to present conclusions in an explicit
format.
In summary the antibacterial activity is mainly determined by
the release of alkaline ions to the medium and an increase of pH.
Important variations in BG antibacterial activity on different
species of bacteria are mainly linked to the composition of BG and,
in particular, to their content of calcium ions. The differences found
in the susceptibility of different bacterial species with respect to
this chemical entity, should lead to addressing the study of the
antibacterial activity of BGs on a wide selection of bacterial ora
to the possible impact on the pathology or the process in which its
use will be involved.
Microbiological conditions studied (culture and coculture media
and time) do not have a signicant impact on the results of the
studies of BGs antimicrobial activity as soon as they are suitable for
the culture of the species under investigation.
Conicts of interest
Authors declare no conicts of interest.
Acknowledgements
Authors thank Mrs. J. Menis for her help in the correction
of the English version of the work. This work was supported
by the POCTEP 0330IBEROMARE1P project, FEDER. ML (PR2010-
0460) thanks the Spanish Ministry of Education for her nancial
support during their sabbatical year at Faculty of Science, University
of Utrecht, The Netherlands.
 M.M. Echezarreta-Lpez, M. Landin / International Journal of Pharmaceutics 453 (2013) 641647
Appendix A. Supplementary data
Supplementary data associated with this article can be
found, in the online version, at http://dx.doi.org/10.1016/j.ijpharm.
2013.06.036.
References
Abou Neel, E., Ahmed, I., Pratten, J., Nazhat, S., Knowles, J., 2005. Characterisation of
antibacterial copper releasing degradable phosphate glass bres. Biomaterials
26, 22472254.
Allan, I., Newman, H., Wilson, M., 2001. Antibacterial activity of particulate Bioglass
against supra- and subgingival bacteria. Biomaterials 22, 16831687.
Balamurugan, A., Balossier, G., Laurent-Maquin, D., Pina, S., Rebelo, A., Faure, J., Fer-
reira, J., 2008. An in vitro biological and anti-bacterial study on a solgel derived
silver-incorporated bioglass system. Dent. Mater. 24, 13431351.
Bellantone, M., Coleman, N.J., Hench, L.L., 2000. Bacteriostatic action of a novel four-
component bioactive glass. J. Biomed. Mater. Res. 51, 484490.
Bellantone, M., Williams, H.D., Hench, L.L., 2002. Broad-spectrum bactericidal activ-
ity of Ag2 O-doped bioactive glass. Antimicrob. Agents Chemother. 46, 1940.
Brauer, D.S., Rssel, C., Kraft, J., 2007. Solubility of glasses in the system
P2 O5 CaOMgONa2 OTiO2 : experimental and modeling using articial neural
networks. J. Non-Cryst. Solids 353, 263270.
Campoccia, D., Montanaro, L., Arciola, C.R., 2006. The signicance of infection
related to orthopedic devices and issues of antibiotic resistance. Biomaterials
27, 23312339.
Colbourn, E.A., Rowe, R.C., 2009. Novel approaches to neural and evolutionary com-
puting in pharmaceutical formulation: challenges and new possibilities. Future
1, 713726.
Day, R.M., Boccaccini, A.R., 2005. Effect of particulate bioactive glasses on human
macrophages and monocytes in vitro. J. Biomed. Mater. Res. A 73, 7379.
Domingues, Z., Corts, M., Gomes, T., Diniz, H., Freitas, C., Gomes, J., Faria, A., Sin-
isterra, R., 2004. Bioactive glass as a drug delivery system of tetracycline and
tetracycline associated with [beta]-cyclodextrin. Biomaterials 25, 327333.
Elliott, A., 2001. Recent developments in non-excitable cell calcium entry. Cell Cal-
cium 30, 7393.
Estrela, C., Bammann, L., Pimenta, F., Pcora, J., 2001. Control of microorganisms
in vitro by calcium hydroxide pastes. Int. Endod. J. 34, 341345.
Froum, S., Cho, S.C., Rosenberg, E., Rohrer, M., Tarnow, D., 2002. Histological
comparison of healing extraction sockets implanted with bioactive glass or dem-
ineralized freeze-dried bone allograft: a pilot study. J. Periodontol. 73, 94102.
Gallego, P.P., Gago, J., Landn, M., 2011. Articial Neural Networks Tech-
nology to Model and Predict Plant Biology Process. Articial Neural
NetworksMethodological Advances and Biomedical Applications. Intech Open
Access Publisher, Croatia, pp. 197216.
Garca-Infante, J.C., Medel-Jurez, J.J., Snchez-Garca, J.C., 2010. Evolutive neural
net fuzzy ltering: basic description. J. Intell. Learn. Syst. Appl. 2, 1218.
Garcia-Domenech, R., de Julian-Ortiz, J., 1998. Antimicrobial activity characteriza-
tion in a heterogeneous group of compounds. J. Chem. Inf. Comput. Sci. 38,
445449.
Gerhardt, L.C., Boccaccini, A.R., 2010. Bioactive glass and glass-ceramic scaffolds for
bone tissue engineering. Materials 3, 38673910.
Gorriti, M.F., Lpez, J.M.P., Boccaccini, A.R., Audisio, C., Gorustovich, A.A., 2009.
In vitro study of the antibacterial activity of bioactive glass-ceramic scaffolds.
Adv. Eng. Mater. 11, B67B70.
Gristina, A.G., 1987. Biomaterial-centered infection: microbial adhesion versus tis-
sue integration. Science 237, 1588.
Hench, L.L., 1998a. Bioactive materials: the potential for tissue regeneration. J.
Biomed. Mater. Res. 41, 511518.
Hench, L.L., 1998b. Biomaterials: a forecast for the future. Biomaterials 19,
14191423.
Hench, L.L., Paschall, H., 1973. Direct chemical bond of bioactive glass-ceramic mate-
rials to bone and muscle. J. Biomed. Mater. Res. 7, 2542.
Hench, L.L., Splinter, R., Allen, W., Greenlee, T., 1971. Bonding mechanisms at the
interface of ceramic prosthetic materials. J. Biomed. Mater. Res. 5, 117141.
Hu, S., Chang, J., Liu, M., Ning, C., 2009. Study on antibacterial effect of 45S5 Bioglass .
J. Mater. Sci. Mater. Med. 20, 281286.
Jimnez, R., Cervantes, C., 2006. Determinacin de la topologa de transportadores
bacterianos. Revista de Educacin Bioqumica, 311.
Kalmodia, S., Molla, A.R., 2010. In vitro cellular adhesion and antimicrobial property
of SiO sub (2)-MgOAl sub (2) O sub (3)-K sub (2) O 013; B sub (2) O sub (3)-F
glass ceramic. J. Mater. Sci. Mater. Med. 21, 12971309.
Kavli, T., Weyer, E., 1994. IEE Colloquium on Advances in Neural Networks for Control
and Systems , pp. 3/13/7.
Kellomki, M., Niiranen, H., Puumanen, K., Ashammakhi, N., Waris, T., Trml, P.,
2000. Bioabsorbable scaffolds for guided bone regeneration and generation. Bio-
materials 21, 24952505.
Landn, M., Rowe, R., York, P., 2009. Advantages of neurofuzzy logic against conven-
tional experimental design and statistical analysis in studying and developing
direct compression formulations. Eur. J. Pharm. Sci. 38, 325331.
Leppranta, O., Vaahtio, M., Peltola, T., Zhang, D., Hupa, L., Hupa, M., Ylnen, H., Salo-
nen, J.I., Viljanen, M.K., Eerola, E., 2008. Antibacterial effect of bioactive glasses
647
on clinically important anaerobic bacteria in vitro. J. Mater. Sci. Mater. Med. 19,
547551.
McHugh, C.P., Zhang, P., Michalek, S., Eleazer, P.D., 2004. pH required to kill Entero-
coccus faecalis in vitro. J. Endod. 30, 218219.
Misra, S.K., Ansari, T.I., Valappil, S.P., Mohn, D., Philip, S.E., Stark, W.J., Roy, I.,
Knowles, J.C., Salih, V., Boccaccini, A.R., 2010. Poly(3-hydroxybutyrate) multi-
functional composite scaffolds for tissue engineering applications. Biomaterials
31, 28062815.
Mortazavi, V., Nahrkhalaji, M.M., Fathi, M., Mousavi, S., Esfahani, B.N., 2010. Antibac-
terial effects of solgel-derived bioactive glass nanoparticle on aerobic bacteria.
J. Biomed. Mater. Res. A 94, 160168.
Mulligan, A., Wilson, M., Knowles, J., 2003. The effect of increasing copper content in
phosphate-based glasses on biolms of Streptococcus sanguis. Biomaterials 24,
17971807.
Munukka, E., Leppranta, O., Korkeamki, M., Vaahtio, M., Peltola, T., Zhang, D., Hupa,
L., Ylnen, H., Salonen, J.I., Viljanen, M.K., 2008. Bactericidal effects of bioactive
glasses on clinically important aerobic bacteria. J. Mater. Sci. Mater. Med. 19,
2732.
Nakajo, K., Komori, R., Ishikawa, S., Ueno, T., Suzuki, Y., Iwami, Y., Takahashi, N., 2006.
Resistance to acidic and alkaline environments in the endodontic pathogen
Enterococcus faecalis. Oral Microbiol. Immunol. 21, 283288.
Nikaido, H., 2003. Molecular basis of bacterial outer membrane permeability revis-
ited. Microbiol. Mol. Biol. Rev. 67, 593656.
Nikaido, H., 1993. Transport across the bacterial outer membrane. J. Bioenerg.
Biomembr. 25, 581589.
Norris, V., Grant, S., Freestone, P., Canvin, J., Sheikh, F., Toth, I., Trinei, M.,
Modha, K., Norman, R.I., 1996. Calcium signalling in bacteria. J. Bacteriol.
178, 3677.
Onoda, T., Oshima, A., Fukunaga, N., Nakatani, A., 1992. Effect of Ca2 and K on the
intracellular pH of an Escherichia coli L-form. J. Gen. Microbiol. 138, 1265.
Peltola, T., Jokinen, M., Rahiala, H., Levnen, E., Rosenholm, J., Kangasniemi, I., Yli-
Urpo, A., 1999. Calcium phosphate formation on porous solgel-derived SiO2
and CaOP2 O5 SiO2 substrates in vitro. J. Biomed. Mater. Res. 44, 1221.
Pye, A., Lockhart, D., Dawson, M., Murray, C., Smith, A., 2009. A review of dental
implants and infection. J. Hosp. Infect. 72, 104110.
Roth, W.G., Leckie, M.P., Dietzler, D.N., 1985. Osmotic stress drastically inhibits active
transport of carbohydrates by Escherichia coli. Biochem. Biophys. Res. Commun.
126, 434441.
Rowe, R., Colbourn, E.A., 2003. Neural computing in product formulation. Chem.
Educ. 8, 18.
Rowe, R., Landn, M., 2013. Articial neural networks technology to model, under-
stand and optimize drug formulations. In: Aguilar Diaz, J.E. (Ed.), Formulation
Tools for Pharmaceutical Development. Biohealthcare Publishing Ltd., Baldock,
pp. 1040.
Shao, Q., Rowe, R.C., York, P., 2006. Comparison of neurofuzzy logic and neural
networks in modelling experimental data of an immediate release tablet for-
mulation. Eur. J. Pharm. Sci. 28, 394404.
Siqueira Jr., J.F., de Uzeda, M., 1997. Intracanal medicaments: evaluation of the
antibacterial effects of chlorhexidine, metronidazole, and calcium hydroxide
associated with three vehicles. J. Endod. 23, 167169.
Stoor, P., Sderling, E., Grenman, R., 1999. Interactions between the bioactive glass
S53P4 and the atrophic rhinitis-associated microorganism Klebsiella ozaenae. J.
Biomed. Mater. Res. 48, 869874.
Stoor, P., Sderling, E., Salonen, J.I., 1998. Antibacterial effects of a bioactive glass
paste on oral microorganisms. Acta Odontol. 56, 161165.
Sun, J.Y., Yang, Y.S., Zhong, J., Greenspan, D.C., 2007. The effect of the ionic products
of Bioglass dissolution on human osteoblasts growth cycle in vitro. J. Tissue
Eng. Regen. Med. 1, 281286.
Vaahtio, M., Munukka, E., Leppranta, O., Zhang, D., Eerola, E., Ylnen, H.O., Peltola, T.,
2006. Effect of ion release on antibacterial activity of melt-derived and solgel-
derived reactive ceramics. Key Eng. Mater. 309, 349354.
Vern, E., Ferraris, S., Miola, M., Fucale, G., Maina, G., Martinasso, G., Canuto, R., Di
Nunzio, S., Vitale-Brovarone, C., 2008. Synthesis and characterisation of bioac-
tive and antibacterial glassceramic. Part 1. Microstructure, properties and
biological behaviour. Adv. Appl. Ceram. 107, 234244.
Verrier, S., Blaker, J.J., Maquet, V., Hench, L.L., Boccaccini, A.R., 2004. PDLLA/Bioglass
composites for soft-tissue and hard-tissue engineering: an in vitro cell biology
assessment. Biomaterials 25, 30133021.
Vogel, M., Voigt, C., Gross, U.M., Mller-Mai, C.M., 2001. In vivo comparison of bioac-
tive glass particles in rabbits. Biomaterials 22, 357362.
Waltimo, T., Brunner, T., Vollenweider, M., Stark, W., Zehnder, M., 2007. Antimicro-
bial effect of nanometric bioactive glass 45S5. J. Dent. Res. 86, 754.
Waltimo, T., Mohn, D., Paqu, F., Brunner, T., Stark, W., Imfeld, T., Schtzle, M., Zehn-
der, M., 2009. Fine-tuning of bioactive glass for root canal disinfection. J. Dent.
Res. 88, 235.
Xie, Z.P., Zhang, C.Q., Yi, C.Q., Qiu, J.J., Wang, J.Q., Zhou, J., 2009. In vivo study effect
of particulate Bioglass in the prevention of infection in open fracture xation.
J. Biomed. Mater. Res. B: Appl. Biomater. 90, 195201.
Zehnder, M., Waltimo, T., Sener, B., Soderling, E., 2006. Dentin enhances the effec-
tiveness of bioactive glass S53P4 against a strain of Enterococcus faecalis. Oral
Surg. Oral Med. Oral Pathol. Oral Radiol. Endodontol. 101, 530535.
Zhang, D., Leppranta, O., Munukka, E., Ylnen, H., Viljanen, M.K., Eerola, E., Hupa,
M., Hupa, L., 2010. Antibacterial effects and dissolution behavior of six bioactive
glasses. J. Biomed. Mater. Res. A 93, 475483.